Anthropic adds Mononoke credit to Scion post after former Meta engineer objects

Anthropic credits Mononoke only after pushback

Anthropic has added a footnote acknowledging Meta's Mononoke to its blog post about Scion, the version control system it built to handle merges from AI agents at scale. The credit came only after a former Meta engineer publicly argued that Scion is a near-complete rebuild of Mononoke, the version control server she helped design.

According to an account published on dev.to on October 16, 2026, Anthropic's original post presented Scion as an AI-era answer to version control: a system able to absorb roughly 30,000 code merges per day from AI agents, built in about five months. The post credited Meta's open-source Sapling client for "hard-won lessons" but made no mention of Mononoke, the server-side system that Sapling was built to work with.

The claimed resemblance

The engineer, identified in the post as Rain, helped design Mononoke at Meta in 2018. As reported by dev.to, she recognised her own work in Anthropic's write-up: the blog used the same "derived data" terminology she had used in materials prepared for an Oxide job application in 2022, framed the architecture the same way, and included diagrams that side-by-side comparisons showed to be nearly identical to Mononoke's.

The claimed overlap goes beyond vocabulary. Per the post, it extends to the overall data model, the Bonsai changeset concept at the core of Mononoke, the server-side API, and a set of architectural decisions considered contrarian when Meta adopted them. The post also raises, without proof, the possibility that the name "Scion" plays on "Bonsai," calling the syllable reversal perhaps coincidental but, given everything else, probably not.

From benefit of the doubt to claimed confirmation

Rain's first reaction was charitable, the post says. Mononoke is open source and present in most training corpora, so Anthropic's models might have converged on the same design independently. She then says she received private confirmation that the resemblance was intentional, and that people familiar with Mononoke's design worked on Scion. Neither claim is independently verifiable, and the dev.to post is the sole source for them.

Her position, in her own words as quoted by dev.to: "I am completely fine with good ideas being copied. I am not fine with the serial numbers being accidentally stripped off, and I am certainly not fine with them being intentionally stripped off."

Anthropic's response

After Rain's posts gained traction — dev.to notes 230 points on Hacker News for the follow-up alone — Anthropic added a footnote acknowledging Mononoke's influence, and an Anthropic staffer named Nick G. responded on Hacker News. The acknowledgement has not settled the dispute. Rain's assessment, as quoted: "This remains yet another shameful example of large AI companies appropriating the commons and presenting others' work as their own." The post also quotes her as saying she now defaults to assuming Anthropic's public engineering presentations are "largely bad-faith ripoffs."

Why it matters

The dispute is less about whether ideas were copied than about how AI-assisted development handles credit. As the post argues, an LLM reproducing a design from its training data has no built-in mechanism to cite where that design came from. A human engineer adapting Mononoke would naturally mention the influence; a model simply outputs the architecture, stripped of provenance.

That also reframes the achievement itself. "Our AI agents built a version control system in five months" is a very different claim from "a team spent years productionising a contrarian design, and we rebuilt it in months by pointing an LLM at the problem." Both can be impressive; only one credits the original work.

The part most likely to recur is Rain's forecast, quoted in the post: "As LLMs continue to assist in designing new systems, they are likely to index heavily on existing designs." If attribution only arrives after the original authors notice and object, most cases will never surface at all.